Kunio Oda

Development of a Low-altitude Hyperspectral Imaging System for Measuring Ground Truth in Agricultural Fields

In the hyperspectral analyses over large agricultural fields, it is necessary to utilize the data acquired by high-altitude platform such as satellites or airplanes. However, in the preprocessing stage of the analyses, it is important to provide with high signal-to-noise ratio (SNR) data acquired on the ground surface, to indicate the ground-truth. In order to obtain the ground-truth hyperspectral data constantly, a low-altitude data acquisition system using a cargo crane with an artificial illumination source is developed. The system can collect data in actual agricultural fields without being seriously affected by the weather condition. Various types of validation data are collected by the system and compared in order to evaluate the performance of the system.

Salt-damaged paddy fields analyses using high-spatial-resolution hyperspectral imaging system

In agricultural fields, the damage caused by salinized winds is crucial for crops. In order to minimize the damage, it is required to detect the damaged areas and enact the proper procedures in order to save the damaged fields immediately after the disaster. In this paper, we propose indices that can indicate the degree of salt-breezed damages in the early withering-up stage. To detect the indices, high-spatial-resolution hyperspectral data taken in actual damaged paddy fields are analyzed. In addition, the sequential change of hyperspectral data in rice within artificial withering-up experiments is recorded to interpret the fundamental mechanism of the indices. The applicability of the indices for satellite data is also shown by applying them to simulated satellite data. Keywords-component; spectroscopy; rice paddy; hyperspectral; salt-damaged; SPOT5; NDVI; NDGI; withering up;

Low-altitude hyperspectral observation of paddy using radio-controlled helicopter

To monitor the chlorophyll content of rice leaves, transparent leave-color monitoring device SPAD; named for the Soil & Plant Analyzer Development, is widely used in Japan. For wide area of paddy, however, it requires lots of elaboration to measure the SPAD values by hand. The objective of this study is to solve this problem by estimating the SPAD values through remote sensing. In rainy season, it is difficult to make use of airborne or satellite image sensors to observe paddy fields. Thus, we made use of a hyperspectral sensor mounted on a low-altitude radio-controlled helicopter, and calculated new indices LVIpure (Leave-color Verified Index) and LVImix using the obtained data to estimate the SPAD values. As a result, the relations between the SPAD value and the estimation index exhibit high correlation: R2=0.885 for LVIpure and R2=0.927 for LVImix, which facilitated to obtain “the estimated SPAD value image” from the hyperspectral data.

Prediction of Sweetness and Nitrogen Content in Soybean Crops from High Resolution Hyperspectral Imagery

In this paper, we investigate a hyperspectral imagery data processing method to predict the sweetness and amino acid content of green vegetal soybean crops. Regression models based on neural networks were developed in order to calculate the level of sucrose, glucose, and nitrogen concentration, which can be related to sweetness and amino acid concentration of vegetables. We demonstrate the method using hyperspectral data of wavelengths ranging from the visible to the near infrared acquired from an experimental field of green vegetal soybeans. A performance analysis is reported comparing regression models built using datasets pre-processed using the first and second derivative analysis. The second derivative transformed dataset presented the best performance overall. Glucose could be predicted with greater accuracy.

Hyperspectral Image Classification of Grass Species in Northeast Japan

This paper investigates the application of artificial neural networks for classifying grass species from hyperspectral image data. High-resolution spatial and spectral data of localized fields were collected using a hyperspectral sensor mounted on the tip of a crane. The hyperspectral datasets are processed using normalization and second derivative in order to reduce the effect of variations in the intensity level of reflectance and to improve the classification accuracy and generalization performance of the neural network-based model. An experimental comparison of the pre-processing methods shows that the best classification accuracy is obtained by the second derivative transformed dataset. Normalization, and a combination of both methods, did not improve accuracy of the neural network models of our experimental datasets more than simple raw reflectance.

Analysis of salt-damaged paddy field using SPOT5 satellite images in Yamagata Prefecture

Abstract —In this paper we propose a new method for estimating the degree of salt-damage over a widespread area using SPOT5 satellite images. In the method, the paddy field pixel data are normalized and plotted in the spectral feature space of NDVI (Normalized Differential Vegetation Index) and NDGI (Normalized Differential Green Index). In the feature space, normal paddies clustered along a standard line, whereas the damaged paddies, at an early stage, shifts at the under side of the standard line. Thus the degree of early-stage damage can be detected from the difference from the standard line. The resultant degree of damages are shown in color and plotted on the map. In order to comparatively evaluate our method, we also examined other methods using reflectance of NIR or NDVI. In the comparison, we confirmed that the newly proposed method showed the best match to the ground truth observation. Subsequently, we attempted to apply the method to the images taken after “Tsunami” in Sumatera for testing the availability.

Evaluation of paddy yield and protein estimation methods based on various vegetation indices, NDSI and PLS using an airborne hyperspectral sensor AISA in Shonai Plain, Yamagata, Japan

This paper describes to evaluate rice yield and protein estimation methods based on various vegetation indices (VIs), NDSI and PLS using an airborne hyperspectral sensor AISA in Shonai plane, northeast Japan. In several developing stages, which are the tillering stage (middle June), the maximum tiller number stage (early July) and the dough ripe stage (late August), PLS has stable and high correlation for all stages. NDSI shows several discriminative wavelengths to estimate rice conditions. VIs slightly estimated those situations in the dough ripe stage. In this result, a hybrid method of PLS and NDSI, which is similar to iPLS, suggests the best estimation method for rice yield and protein.

Leaf area index estimation from hyperspectral data using group division method

In this study, optical remote sensing data were used for leaf area index (LAI) estimation. The LAI is an important measure to increase the yield and adjust the quantity of manure. LAI extracted from remotely sensed data may contribute to grasp the yield of rice at an early stage. Therefore, the purpose of this study is to estimate the LAI through remote sensing. For the purpose of our work, we proposed a Group Division Method. The method can decide a set of optical bands for estimating the value of LAI. The index is extracted by comparing the order of ground truth data with that of spectral data. As a result, the effective index to estimate LAI is made by reflectance relations in 545nm, 1170nm and 1290nm from the hyper-spectral data of rice field in Sakata City, Yamagata prefecture. Furthermore, we applied the index to the data set obtained in Furukawa, Miyagi prefecture to verify the effectiveness of the method. Finally the “LAI estimate map” was made and examined whether this study helped an automatic LAI estimate in wide area.

Monitoring of Bacterial Pustule on Soybean by Neural Network Using Hyperspectral Data

This paper proposes a technique for estimating the soybean bacterial pustule infection using a neural network, provided with hyperspectral data obtained from the cranemounted hyperspectral sensor system. This technique uses the concept of a quotient set in manifold to reduce the feature space of hyperspectral data. This preliminary study shows that the manifold-embedded neural network is effective to classify bacterial pustule degrees in the nonlinearly reduced hyperspectral feature space. Keywords-bacterial pastule; feature map; hyperspectral data; manifold; neural network; quotient set; soybean